Exa Places: search local businesses and verify the returned data

Use Exa Places to retrieve candidate local businesses, then verify that each result refers to the intended branch and supplies evidence for the fields your task needs. The current Search API reference provides the request contract. The request below is an illustrative configuration, not a completed API call.

Make the location and task explicit

POST https://api.exa.ai/search
Content-Type: application/json
x-api-key: YOUR_EXA_API_KEY

{
  "query": "Independent bookstores in Brooklyn, New York",
  "type": "lite",
  "numResults": 5,
  "userLocation": {"country": "US"}
}

Keep the key in a secret store or environment variable; the placeholder is not a credential. The query names the city because a country setting alone does not identify a neighborhood or branch. The current reference accepts the shown fields for Lite. It does not define a dedicated places category, so the example does not invent one.

Lite serves content from Exa’s refreshed search index without fetching pages at query time. The Places announcement now recommends Lite for low-latency search while identifying Fast as the mode used in its earlier benchmark. That recommendation does not turn the old benchmark into a new Lite measurement.

Map the response to a verification record

Preserve the raw response and retrieval time before extracting a business row. Match the returned identity and source URL to the requested place. If a field such as hours, operating status, or telephone number is absent or not currently supported, keep it unresolved rather than filling it from a nearby business.

Download the local-business verification record (CSV)

The record separates query, requested locality, returned business, branch identity, source URL, field, observed value, checked time, and verification decision. Remove its fictional example before use. A retrieval timestamp is not proof that the underlying business field was updated then.

SearchEngineAnswer checked the request documentation but made no paid Exa call and has not measured the example’s results. The original branch and field checks below remain the acceptance criteria for any later live sample.

An assistant recommending a shop open tonight needs more than a plausible business name. It needs the correct branch, a current operating status, and hours applicable to the requested date. Exa’s new Places data makes those fields easier for agents to consume, but a structured answer still needs verification, since getting it wrong could send someone to the wrong place.

In its October 5 announcement, Exa reports adding 211 million places worldwide across its endpoints at no additional Places charge. Listed fields include business names, categories, addresses, coordinates, operating status, hours, phone numbers, websites, and ratings. The total is Exa’s claim, not an independently verified count of complete records across all countries.

Resolve identity before measuring a discrepancy

Start with a fictional business, Harbor Books, which has two branches. A returned record says it closes at 8 p.m.; a website page says 6 p.m. Before calling either source stale, compare the address, coordinates, branch name, and website destination. Two correct records for different branches can appear as a single incorrect record when the audit matches on name alone.

If both refer to the same branch, check the date and the kind of hours. Regular hours, holiday hours, and a temporary closure can all be valid in different contexts. Do not average 6 p.m. and 8 p.m. into a new 7 p.m. closing time. The disagreement is a reason to seek a more specific record or qualify the recommendation.

Attach evidence to the field

An address and a phone number can have different freshness histories. Saving one timestamp for the whole business hides that distinction. For a small audit, attach the returned value, official-page value, retrieval time, and any displayed effective date to each important field.

• Identity: Does the record refer to the intended branch rather than the parent business?
• Hours: Are these regular hours or an exception for the requested day?
• Website: Does the destination belong to the business and resolve to the relevant location?
• Phone: Is it a branch number, central switchboard, or an unrelated record?
• Status: Is a temporary closure distinguishable from a permanent one?

Mark a field unknown when you cannot settle it. Missing evidence is not the same as a verified mismatch. A business-owned page is a useful source, but if it has no effective date or is internally inconsistent, report that limitation instead of treating ownership as proof of current accuracy. For an immediate visit, a direct confirmation may be more useful than another directory result.

The vendor benchmark answers a narrower question

Exa also describes a 400-query evaluation using F1 for businesses returned with correct metadata. That is a vendor-run retrieval evaluation. It is not a test of this publisher’s local-business sample, proof of Google Maps ranking, or evidence that every returned hour is current.

A useful buyer comparison would hold the business sample, query wording, geography, and evaluation date constant. It would publish identity errors separately from field errors. Returning the wrong branch, even with impeccable metadata, shouldn’t count as a successful answer to a request about another branch.

How local publishers can use the change?

For a local guide or directory, use structured data to locate candidates, then verify the fields your recommendation depends on. A “best places” article needs editorial reasons for inclusion; a business dataset does not supply your judgment. A route planner needs current practical details; a high rating does not settle opening hours.

Our local recommendation comparison separates source discovery from the final recommendation. The benchmark explainer shows why finding the right set and completing every record are different evaluation jobs. This report describes the announced capability and a review method; no independent Places API accuracy test has been completed here.

For the separate October 19 retrieval release, the Exa 3.0 evaluation guide compares Lite and Auto configuration choices while keeping source support and total cost explicit.